Event Participants and Implicit Arguments Experimental Approaches to Verb Meaning
GIVE TAKE/STEAL FRIGHTEN GIVE location TAKE recipient agent theme experiencer time STEAL victim FRIGHTEN direction of transfer transfer
Thematic Roles and Event Participants Agent Individual who performs/does the action of the event Patient Individual who undergoes/is affected by the action of the event Instrument Individual that assists with the action of the event Source Where the action initiates Goal Where the action is going Location Where the event takes place Time When the event takes place Experiencer Individual who comes to be in a mental states Benefactive Individual who the event is done for.
Thematic Roles and Event Participants λxλe[ Agent(e, x) ] = def λxλe[ e [ Do(e, x) & Cause(e, e) ]] λxλe[ Patient(e, x) ] = def λxλe[ e [ Become(e, e) & Result(e, x) ]] λxλe[ Instrument(e, x) ] = def λxλe[ e [ Assist(e, x) & Cause(e, e) ]]
Aside on Instruments Why not just have λxλe[ e [ Assist(e, x) ]? John broke the window with a stone. The stone broke the window. The window was broken with a stone. *The window broke with a stone. (Anticausitive)
IMPLICIT ARGUMENTS
Agents The collector sold the vase immediately. The vase was sold immediately. The vase was sold by the collector immediately. The vase sold immediately. *The vase sold by the collector immediately.
Arguments: Entailed or Implicit? Event participants can be entailed #The collector sold the vase immediately, yet no one had sold it. #The vase was sold immediately, yet no one had sold it. #The vase sold immediately, yet no one had sold it. #The wheel was spun, but no one spun it. The wheel spun, but no one spun it. But only some are implicit The collector sold the vase immediately to raise money for charity. The vase was sold immediately to raise money for charity. #The vase sold immediately to raise money for charity. The wheel was spun to see what prize was won. #The wheel spun to see what prize was won.
The donated antique vase had sold immediately, yet no one had sold it. Intransitive The donated antique vase was sold immediately, yet no one had sold it. Short Passive Rachael is a strict vegetarian so she eats prime rib for dinner every day. Non-Agent Contradiction Max is incompetent, so we hired him. Implausible Mauner & Koenig (2011)
% Judged Contradictory 100 90 80 70 60 50 40 30 20 10 0 Implausible Contradictory Intransitive Short Passive Mauner & Koenig (2011)
Summary Verbs like sell entail agents. Detection rates for agent contradiction is the same in short passives and in intransitives. Similar to more overt contradictions Dissimilar from mere implausibility So are implicit agents the same as entailed agents? Mauner & Koenig (2011)
The donated antique vase was sold immediately to raise some money for the charity. Short Passive The donated antique vase had sold immediately to raise some money for the charity. Intransitive Mauner & Koenig (2011)
Stop-Making-Sense Judgments + The vase had sold immediately to + The vase was sold immediately to raise raise some some money money Mauner & Koenig (2011)
* * * * Mauner & Koenig (2011)
Mauner & Koenig (2011)
Summary Participants rapidly detect violations of rationale clauses in intransitives over short passives. Increased cumulative rejection of the rationale clause making sense starting at the verb. Even when participants were not rejecting the sentence, word-by-word judgment times within the rationale clause were also inflated.
To prevent the champ from killing his rival, the fight was ended in the fifth. Short Passive To prevent the champ from killing his rival, the fight had ended in the fifth. Intransitive Mauner & Koenig (2011)
* * * * Mauner & Koenig (2011)
Mauner & Koenig (2011)
Summary Preposed rationale clauses lead participants to reject intransitive continuations. This effect emerged at the verb. Some evidence for inflated reading times for intransitive clauses. What about other types of event participants?
Aside: Coercion vs. Rationale Aspectual Coercion After several minutes, the rabbit jumped. For several minutes, the rabbit jumped. Both are acceptable. But there is a cost to the second. Implicit Agents To prevent the champ from killing his rival, the fight was ended in the fifth. To prevent the champ from killing his rival, the fight had ended in the fifth. Only the first is acceptable. And there is a cost to the second.
Accommodation and Implicit Themes No Implicit Theme Implicit Theme Unloaded what!? Bill hurried to catch his plane. Bill hurried to unload his car. The suitcases were very heavy. The suitcases were very heavy.
100% 1900 95% 1850 1800 90% 85% 1750 1700 1650 Implicit Theme No Implicit Theme Implicit Theme No Implicit Theme 80% 1600 1550 75% Accuracy 1500 RT Final Sentence Judgments
Summary Implicit arguments may guide discourse processes. Accommodation of a definite related to an implicit participant, here an implicit theme, may be eased.
Sluicing The secretary typed something, but I don t know what exactly. the secretary typed The secretary typed, but I don t know what exactly. the secretary typed
Sluicing The secretary typed, but I don t know what exactly. typed what!? The secretary typed, but I don t know where exactly. typed what!?
The secretary wrote something, but nobody seems to remember what exactly. The secretary wrote quickly, but nobody seems to remember what exactly. The secretary wrote with something, but nobody seems to remember with what exactly. The secretary wrote quickly, but nobody seems to remember with what exactly. Argument, Overt Argument, Null Adjunct, Overt Adjunct, Null
1200 Overt vs. Null: F 1 (1,39)=10.25, p<.01; F 2 (1,31)=4.60, p<.05 Argument vs. Adjunct: F 1 =23.01, p<.001; F 2 =8.47, p<.01 1100 1000 900 800 Argument, Overt Argument, Null Adjunct, Overt Adjunct, Null 700 600 subj VP but subj v wh adv
1. ate something/with something 2. write something/with something 3. cooked something/with something 4. type something/somewhere 5. cleaned something/somewhere 6. paint something/on something 7. telephoned someone /from somewhere 8. push something/towards something 9. bake something/for some reason 10. dream something/for some reason 11. screamed something/for something 12. moan something/about something 13. curse something/about something 14. sang something/about something 15. mumble something/about something 16. check something/about something Instrument (with what) Location (where) Reason (why) Indirect Theme? (for/about what)
Summary No clear interaction as expected for an nontheme arguments. But the stimuli are questionable in terms of the status of non-theme implicit arguments. Can we find online evidence for nonagent/non-theme event participants?
Lexical Encoding Hypothesis A participant role is a (semantic) argument of a verb if and only if 1. Its presence is required of all situations described by that verb, and 2. It is required of the denotation of only a restricted set of verbs.
Which sword did the rebels kill the traitor king with during the rebellion? Which sword did the rebels behead the traitor king with during the rebellion? With which sword did the rebels kill the traitor king during the rebellion? With which sword did the rebels behead the traitor king during the rebellion? Instrument permitted, wh-np Instrument required, wh-np Instrument permitted, wh-pp Instrument required, wh-pp
950 900 850 800 750 700 650 600 behead wh-np behead wh-pp kill wh-np kill wh-pp 550 500 Subject Verb DO (with) Raw reading times
~80 msec ~48 msec
Summary The integration cost of a wh-filler is eased when that filler is an event participant.
General Summary Implicit arguments are represented and processed in real-time Implicit agents are detectable with rationale clauses. Implicit themes ease accommodation Implicit instruments ease integration of wh-fillers Question: But how do we know what event participants are privileged?
ARGUMENTS AND EVENT CONCEPTS
Preconditions for Implicit Arguments Argument Condition: The verb must encode more arguments than it necessarily must have overtly expressed.
The rabbit biffed the cookie! Could biff mean take? steal?
The rabbit biffed the cookie from the baby
Preconditions for Implicit Arguments Argument Condition: The verb must encode more arguments than it necessarily must have overtly expressed. Event Concept Condition: The event must be able to be viewed as having certain privileged participants.
GIVE STEAL obligatory arguments THEME AGENT STEAL obligatory arguments VICTIM implicit argument
PICK-UP STEAL AGENT THEME PICK-UP obligatory arguments THEME AGENT STEAL obligatory arguments VICTIM implicit argument
Participant to Argument Match Acquisition/Learning Heuristic: In basic clauses, the nominal satellites of a verb correspond exactly to the verb s event participants. A clause with n arguments expresses a concept with n participants.
The rabbit biffed the cookie Participant-to-Argument Match biff must be a 2-participant concept. This would rule out 3-participant event meanings: X gave Y to Z X put Y in Z X stole Y from Z X baked Y for Z X opened Y with Z X passed Y to Z X sent Y to Z X read Y to Z Argument-to-Participant Match Each argument must be matched with an event participant. 3-participant concepts can be viewed as 2-participant concepts with an implicit third argument. 2+1-participant event concept biff must be a 2+ participant concept.
2- or 3-Participant Concepts? If any event that can be viewed under an n+mparticipant concept is equally likely to be viewed under an n-participant concept, the participant-to-argument matching cannot narrow down the meaning of a verb (Williams 2015).
How do we know whether this event was viewed under a 2-participant or 3-participant concept?
Similarity Judgment Task
4 Experiments 1. Give (3-participant) vs. Hug (2-participant) Initial control study: Obligatory arguments 2. Hit (2-participant) vs. Bean (2+1-participant) 3. Pick-up (2-participant) vs. Steal (2+1-participant) 4. Open (2-participant) vs. Jimmy (2+1-participant) Three Conditions: Token videos are drawn from the same event type. Perceptual videos are drawn from the same event type, but the action direction is changed. Critical One video is drawn from each type of event.
Give vs. Hug Give: A girl gives another girl a teddy bear. Hug: A girl hugs another girl while holding a teddy bear.
Give (3-participant) vs. Hug (2- participant) Similarity Rating Response Time
Summary Subjects were sensitive to the number of (obligatory) participants The effect came out of response times instead of similarity ratings themselves. Critical comparisons took longer to rate than other same type comparisons. So what about our implicit argument cases?
Pick-up vs. Steal Pick-up: A girl picks up a toy from a table while another girl stands by. Steal: A girl steals a toy from another girl s hands.
Pick-up (2-participant) vs. Steal (2+1- participant) Similarity Rating Response Time
Hit vs. Bean Hit: A girl hits another girl with her hand while holding a blue ball. Bean: A girl beans another girl by throwing a blue ball.
Hit (2-participant) vs. Bean (2+1- participant) Similarity Judgment Response Time
Open vs. Jimmy Open: A girl opens a suitcase while holding a hook. Jimmy: A girl opens a suitcase using a hook.
Open (2-participant) vs. Jimmy (2+1- participant) Similarity Judgment Response Time
Summary In each case, subjects noticed the difference between the critical different type cases and other same type cases. Some 2-argument verbs appear to be related to 3- participant event concepts. So how did the class do?
Practical: Open/Jimmy N = 8 21 Comparisons 2 per event type Similarity Judgments, Correlations Max: 0.94 Average: 0.83 Min: 0.66
Side Change Critical Change
Practical: Open/Jimmy 8 7 6 5 4 3 2 1 0 Similarity Judgment Critical Side Change Token Identity Similarity Judgment
Open, left, 1 Open, left, 2 Open, right, 1 Open, right, 2 Jimmy, right, 1 Jimmy, right, 2 Open, left, 1 7 Open, left, 2 5.875 6.875 Open, right, 1 4.125 4 7 Open, right, 2 4.375 4.625 6.625 6.875 Jimmy, right, 1 2.125 2.375 4 3.875 6.75 Jimmy, right, 2 2 2.25 3.25 4 6.375 7
General Summary Implicit arguments are represented and processed in real-time The event concepts that underlie arguments may create problems for learning. No one-to-one mapping Argument-to-Participants Mapping sets a lower bound for learning. n arguments --> n+m participants.